Feature

From EDN Europe: Navigate the analogue-signal-conditioning maze

Now that you can choose sensors to make virtually any physical measurement, you

By David Marsh, Contributing Technical Editor -- EDN, 12/7/2000

If you work in automotive, industrial or process control, or instrumentation engineering, you know that today's silicon sensors provide reliable and highly stable conversions between most physical quantities and electronics. You also know that there's a growing number of sensor ICs with digital outputs that suit direct connection to digital signal processors and microcontrollers (Reference 1). For low-volume applications, you might select a modular "smart sensor" that includes the signal-conditioning and interface electronics and that may conform to emerging industrial standards (Reference 2). But most often, your customer's application almost exclusively determines your sensor choice, so what happens when you can't choose a ready-made sensor solution or a sensor with a digital output?

Operational amplifiers (op amps) are the analogue designer's glue logic, and they use them in huge numbers; it's now some 25 years since the uA741 revolutionised op-amp design. But digital engineers are often reluctant to use op amps, fearing yesteryear's black magic that was necessary to defeat circuit drift and maintain loop stability. Today's op amps don't suffer from such drawbacks, and their user-friendliness makes them natural for sensor interfaces. If the application fits, you can select dedicated sensor-interface ICs to replace custom op-amp circuits. Such interface ICs can linearise tricky sensor elements such as thermistors, or convert differential voltages from a transducer bridge to drive an ADC. You may also want to consider programmable analogue ICs, with CAD tools that abstract your design and allow you to concentrate on a computational approach to continuous-time electronics.

Whichever approach you favour, under the surface you'll find op-amp cells implementing the amplifier and filter stages you need for sensor interfaces. But anyone who can design a DSP system or code a block of C++ can easily tackle analogue, and at least well enough to connect a typical sensor to the digital comfort zone. The IC vendors think so too, as you'll find from the stream of seminar invitations for DSP engineers who need to connect with real-world physical properties. Uwe Fuhrmann, European strategic marketing manager for Texas Instruments' advanced analogue products, says that his company sees most students (in Universities today) studying software and digital hardware: "Analogue is still growing fast, so we're trying to add value with high-performance products that are also easy to use." Vendors such as TI recognise today's shortage of analogue talent with tutorials and extensive application support on their Web sites, and you'll also find several independent analogue resources.

Todays's op amps banish black magic

Analogue sensor interfaces invariably follow a sensor/amplifier model where the first amplifier stage provides the primary level, offset and scale correction, and a second stage provides Nyquist-frequency filtering to prevent aliasing before the ADC. If you're looking for the fastest conversion times, you may need impedance matching to drive A/D flash converters, and possibly differential drive, too. Op amps are the obvious choice for these tasks, and today's process technology improvements make them predictable and easy to use. Now, your biggest problem is to pick the optimum op amp from the bewildering array from vendors such as Analog Devices, Burr-Brown, Elantec, Linear Technology, Maxim, Micrel, Motorola, National Semiconductor, Philips, and Texas Instruments (Reference 3).

To illustrate a typical sensor/amplifier design, consider the commonest physical measurement: temperature. The cheapest temperature sensors are thermistors or their silicon equivalents—spreading-resistance sensors that rely on the specific resistivity of bulk silicon, such as Infineon's KT/KTY series. You can choose thermistors with positive or negative temperature coefficients (PTC/NTC), while spreading-resistance devices increase their resistance value with rising temperature (PTC). Either way, the sensor requires an excitation current to generate an output voltage that's proportional to temperature. For example, data for Infineon's KT/KTY sensors shows that the output voltage exactly tracks a shallow but regressive parabolic curve. The sensor's value is normalised to either 1 or 2 kW at 25°C and ranges from a factor of 0.518 to 2.235 of that value across the –50 to 150°C operating range. There's about ±2% resistance tolerance at temperature extremes.

It's easiest to compensate the parabolic response and calibrate out tolerances in software, reducing the analogue-design task to biasing, level shifting and filtering. The recommended bias current is 1 mA, and that value must be stable over temperature, requiring a current source. Current-source diodes such as Vishay's J505 typically have ±20% initial tolerances and drift characteristics of about 0.33 %/°C at the 1-mA level. But to a first approximation, the performance of a voltage-reference and op-amp current source mirrors the voltage reference and reference resistor's performance (Figure 1a). Precision resistors cost pennies, while bandgap voltage references such as Analog Devices' ADR391 have temperature coefficients of just 25 ppm/°C over the automotive/industrial temperature range. Operating from 3 to 18V and consuming less than 120 µA, the ADR391 provides a 2.5V output, includes sense and shutdown inputs, and comes in tiny SOT-23 packages for $1.23 (10,000).

To visualise any op-amp circuit's operation, first assume perfect components working within their linear region, and then remember just two points: Feedback forces the op amp's inverting input to mirror the voltage at its noninverting input, and neither input terminal takes any current. So, in Figure 1a, the op amp's output drives to a value that applies 2.5V to the inverting input. All the current passes through the 2k499 reference resistor, so the current through the sensor is constant at 1 mA. Notice that sensing the output voltage directly across the sensor nullifies the op amp's input offset voltage (Vos, the real-world difference voltage between inverting and noninverting inputs) but also means that you'd need a high-impedance amplifier or buffer stage to avoid loading the current source.

Assuming a 1-kW sensor that's grounded at one terminal, a KT/KTY's output voltage ranges from about 0.5 to 2.2V, so you need some scaling to match the typical ADC's input span (say, 0 to 5V). To accommodate working from a single-supply rail, the configuration in Figure 1a floats the sensor, so you also need to restore a ground reference before the ADC's input. Ideally, op amps like to work halfway between the supply rails, or symmetrically around 2.5V for a 5V supply, but you'll often simplify life and improve circuit performance if you can provide a negative supply rail. If you're working from a single supply, be sure to choose op amps that can drive their outputs within a few millivolts of V_CC and ground (rail-to-rail operation). For best linearity over the maximum input range, also choose devices that can accept input voltages close to the rails. For a full discussion on rail-to-rail op amps and single-supply operation, see Reference 4.

Differential- and instrumentation-amplifier ICs simplify design

Restoring Figure 1a's ground reference requires the differential-amplifier stage that typifies sensor input channel front ends. Simple differential amplifiers take the form of Figure 1b, where, providing that R2/R1=R4/R3, V_OUT =V_IN ´(R2/R1). Although it's simple and minimises component count, this circuit requires accurate resistor matching to reject common-mode voltages and has limited input impedance. To improve performance, you might consider single-supply sensor amplifiers, such as Analog Devices' AD22057, or "fully differential" op amps, such as Texas Instruments' THS414x family, which aim at driving differential-input ADCs. Maintaining a fully differential signal path helps to preserve the ADC's input sensitivity by rejecting noise sources, such as power supplies and switching currents (Reference 5). The THS414x family operates with single or dual supplies from 5 to ±15V, comes in an eight-pin SOP and costs about $3.40 (100) for the industrial temperature grade.

One way to restore the differential amplifier's high input impedance uses two op amps (Figure 1c). Here, if R1/R2=R4/R3, V_OUT =V_IN ´(1+R4/R3); notice that the first-stage amplifier limits the input voltage below V_IN ´R1/(R1+R2). The op amps' common-mode input impedance determines the circuit's input impedance (typically anything from 10 MW upwards), and the resistor matching determines common-mode rejection ratio (CMRR). Simplifying design, Burr Brown's INA156 is a recent IC version that's built from CMOS technology to provide a rail-to-rail output swing in single-supply circuits. Demonstrating that analogue CMOS can be precise, the INA156 has a typical input offset voltage of 2.5 mV with an offset voltage drift rate of just ±5 µV/C. Differential and common-mode input impedance is 10¹³ W and the input bias current figure is a maximum ±10 pA. You fix the INA156's gain at 10 or 50 with a link, also setting its bandwidth to 550 or 110 kHz—fast enough for rapidly responding sensors, such as accelerometers. The INA156 comes in miniature eight-pin SOPs for about 95 cents (1000).

But the "proper" way to measure sensitive voltages riding on a common-mode level is to use an instrumentation amplifier. Instrumentation amplifiers traditionally comprise three op amps and six precision resistors, with a seventh resistor setting the circuit's gain (Figure 1d). If R1=R2 and the resistors around the output op amp are all equal, V_OUT =V_IN ´(1+2R1/R7). The two input op amps form a differential buffer with a gain of 1+2R1/R7 for differential signals and unity gain for common-mode signals. Mismatches between R1 and R2 create gain errors but don't affect the circuit's CMRR. Because the circuit's gain is almost always in the front end, it's easy to match R3-R6 (which do affect CMRR) using equal value resistors.

Vendors including Analog Devices, Burr-Brown/Texas Instruments, and Maxim offer conventional, dual-rail precision instrumentation amplifier ICs and new single-rail, low-cost parts for applications such as automotive sensors. One example is Analog Devices' AD623 that works from single- or dual-rail supplies and costs just $1.82 (100). But the low cost doesn't comprise performance, with specifications such as less than 200-µV maximum input offset voltage and ±2-µV/C drift, 25-nA input bias current, and a 1-kHz input noise voltage density of 35 nV/ . One resistor sets the AD623's gain between unity and 1000; for a gain of 10, the device has 100-kHz bandwidth, settles to 0.01% in 20 µsec, and has 90-dB CMRR to line-frequency signals. The device comes in eight-pin packs including DIP, SOIC, and a high-density micro-SOIC with a 6´9.8-mm footprint.

Correct filtering is crucial

The last major element you have to consider is filtering. The minimum filtering that you can get away with limits input frequencies to the ADC to prevent "aliasing." Aliasing occurs when the ADC digitises signals that are higher than half the ADC's sampling frequency (such as noise) to create an ambiguous output signal. For example, if you reconstruct an aliased version of a sine-wave input signal using a DAC, you may see two representations of the input signal—one at the correct frequency and another at an indeterminate frequency. If you have no advance knowledge of the input signal's exact frequency, you can't distinguish the correct representation. To prevent aliasing, you can either raise the ADC's sampling frequency, which is expensive and normally impractical, or include a lowpass filter in the signal chain.

One approach is to use an active filter stage built from one or more op amps. For simple responses such as a second-order Butterworth characteristic and unity gain, all you need is one op amp, two resistors, and two capacitors (Figure 2a). Here, the filter's roll-off is 40 dB/decade, and you can set the cutoff frequency relative to 1 rad/sec (0.16 Hz) simply by scaling the component values. Component matching is easy because the resistors are equal value, and one capacitor is twice the value of its counterpart. You set the circuit's input impedance by choosing the resistor's value, and the cutoff frequency fc is then ½p=[R1´R2´C1´C2]. So, for a cutoff frequency just over 1 kHz with 10-kW input impedance, you might choose 10-kW resistors with a 0.02-µF feedback capacitor and 0.1 µF to ground to provide a 1.125-kHz response (normalising filters to 10-kW levels is an old trick that's more intuitive than 1 rad/sec). You can adjust the passband gain by adding two resistors to the negative feedback path so that G=1+R3/R4. Another feature is that if you can accept a fixed gain of 1.516, you can use equal component values for C1 and C2 as well as R1 and R2; for the same 1-rad/sec normalisation, the values are 1W and 1F.

It's also possible to include active and passive filtering around the amplifier topologies above (Figure 2b). Here, R1, R2, and R3 and C1 and C2 comprise a second-order multiple-feedback lowpass filter that provides similar responses to Figure 2a. But the ratio R2/R1 sets passband gain and complicates design unless you rationalise your design criteria, such as by fixing gain at unity. If you can accept this compromise, you can again scale the cutoff point relative to 1 rad/sec by substituting 1W for all resistor values together with C1=0.471F and C2=2.12F. At the op amp's output, R4 and C3 provide single-pole lowpass filtering to further attenuate noise, whose values you must balance against the ADC's input impedance.

There are numerous filters to explore including linear (active resistor-capacitor) and switched-capacitor topologies that offer simple and complex responses. One good way to check out the possibilities is to use a filter design program such as Linear Technology's FilterCAD, which you can download free from www.linear.com. Burr-Brown/Texas Instruments also have on-line filter design tools to explore together with a wealth of application notes.

If you're working with instrumentation systems, consider ready-made filter modules from specialists such as Kemo that offer precise responses for high-accuracy applications. The company's Managing Director Robert Owen asserts that you still can't beat modules for such measurements: "At the precision end, the filtering world's classic mistake is to select a two-pole Butterworth chip for the antialiasing filter. You'd almost do better to leave it out." The company's 1200 series provide four- or eight-pole responses, including Bessel, Butterworth, linear phase, and models with flat passband and sharp cutoff responses that optimally suit antialiasing applications.

ASSPs solve problems, go digital

Although they're simple and easy to master, the circuit topologies above account for most sensor interfacing needs. You can clean up "random-logic" op-amp designs in an analogue array and use the free configuration software to build and simulate virtual circuits (see Reference 6 and sidebar "Program analogue arrays on the fly" for a description of working with programmable analogue parts from Anadigm and other vendors). But for special circumstances such as inductive, magnetic-field, and pressure measurements, consider application-specific signal processors (ASSPs). Inductive sensors present a tricky interfacing problem when you want to measure rotational speed, even over a moderate speed range. At low speeds, the sensor's output is small, so you have to set a Schmitt trigger or comparator's threshold value accordingly. At higher speeds, the output voltage rises substantially, along with the noise level in automotive applications, causing false triggering if the threshold values are too low. Toshiba avoids this situation with its TA8025 pulse-shaping IC that converts the sensor's sinusoidal output waveform to a logic-compatible rectangular format. A self-biasing circuit adjusts the inductive sensor's output levels to preserve sensitivity and noise immunity over a dc to 50-kHz range. The TA8025 is available in eight-pin DIPs and SOICs for about $1.20 (10,000).

Magnetic-field measurements are set to become far more popular following the introduction of low-cost sensors such as Philips' KMZ41 (about 84 cents (1000)). Such devices convert angular measurements into voltages and can replace wear-prone potentiometers in applications such as throttle control, suspension compression compensation, and even contactless measurements in aggressive fluids—anywhere that you can convert a movement into angular motion, possibly with a simple mechanical linkage. Philips offers the UZZ9000 (analogue output) and UZZ9001 (digital output) to simplify interfacing the KMZ41, for about $2, although you can compute the sensor's major correction coefficients in software. The company will also shortly announce its KMA200, an improved sensor that suits weak field measurements and replaces rare-earth magnetic targets with low-cost ferrous items.

Getting ever closer to the digital comfort zone, Maxim's MAX1460 (Figure 3) is a third-generation signal-processor IC that suits low-level bridge measurements, such as monitoring pressure using piezoresistive microelectromechanical sensors (MEMs). But unlike earlier ASSPs, the MAX1460 includes a digital processor to compensate for a silicon sensor's gain, nonlinearity, and temperature-coefficient variations to better than 0.5% accuracy. The single-rail 5V IC integrates a programmable-gain amplifier and local temperature sensor ahead of a 16-bit ADC to provide coarse gain and temperature coefficient corrections. A 128-bit EEPROM carries the fine correction coefficients that the 16-bit DSP manipulates to provide a 12-bit digital output. The DSP accepts 16 instructions such as add, branch, and multiply from a 128-byte ROM. Alternatively, you can use the chip's DAC to generate an analogue-output voltage and filter it using the spare op amp. The MAX1460 is available in a 48-pin TQFP for about $6.50 (1000), and there's a 2.4V version for portables that draws just 310 µA for the same price (MAX1462). You can also order PC-compatible evaluation kits, one for $250 that includes a sensor and a $200 version with no sensor. Both kits are available now.

New ADCs provide direct connection

Also watch out for new-generation ADCs that can simplify your design tasks by integrating common interface components. Cirrus Logic's industrial ADC range includes the 16- and 20-bit CS551x series and the 24-bit CS553x family that target sensor interfaces. Available now, the CS5510 is an eight-pin SOIC device with a fully differential input and an output word rate of as much as 163 Hz. The 250-mV to 5V voltage-reference range can dispense with the need for amplifiers for sensors with low output levels. The chip includes a digital filter to reject line-frequency interference, has a two-wire serial digital interface, and costs less than $4 (1000). Now available in sample quantities, the CS5534 connects directly to bridge circuits such as strain gauges. A four-channel differential-input multiplexer precedes a chopper-stabilised instrumentation amplifier that you can program for gains of 1 to 64 in powers of two. A digital filter supports conversion rates from 7.5 Hz to 3.84 kHz and outputs results over a three-wire SPI/Microwire-compatible serial interface.

Melexis recently announced its MLX90308 signal-conditioning IC that includes an on-chip microcontroller to perform sensor calibration and linearisation. Optimised for automotive pressure transducer measurements, the IC suits bridge or differential sensor connections. The IC has a local temperature sensor subsystem and an external temperature-sensor input. In the signal path, a 12-bit DAC corrects offsets ahead of a programmable-gain amplifier that normalises signal levels for the chip's 10-bit ADC that can sample at 75 msec intervals. The ADC feeds the custom 8-bit MLX microcontroller core that has 4k ROM and 64 bytes of RAM, together with 384 bits of EEPROM to store calibration data. Corrected signal values pass to a DAC that supplies the output buffer op amp that you configure for voltage or 4- to 20-mA current-loop operation. Software tools are available to program the chip via the IC's PC-compatible serial port. The supply voltage range is 6 to 35V, operating temperature is –40 to +140°C, and the IC comes in a 16-pin wide-body SOIC package. Available now, expect to pay around $4 (10,000), or order an evaluation pc board complete with five sample chips for $100.

Analog Devices will soon release details of its AD7708/18 and AD7719 sensor interface ADCs with glueless interfaces that target relatively low-throughput industrial applications. The 16-bit AD7708 and 24-bit AD7718 handle multiple channels and are pin- and register-compatible. The top-spec AD7719 comprises a 24-bit main channel and a 16-bit secondary channel using dual delta-sigma converters, each capable of 20-Hz conversion rates. The main input channel has differential inputs and an input buffer that works to within 100 mV of ground and the 2.7 to 5.5V positive supply rail. The programmable-gain amplifier and 24-bit ADC combination provides 13- to 18-bit peak-to-peak resolution for input voltages between 20 mV and 2.56V. Each ADC has a separate reference input that accepts voltages from 1V to V_CC , allowing you to implement different scaling factors for each channel. There's also a pair of matched 200-mA current sources to provide sensor stimulus. Expect to see the AD7719 appear in 28-pin SOIC packages and TSSOPs during March/April 2001.